User-Item Matching for Recommendation Fairness

As we all know, users and item-providers are two main parties of participants in recommender systems. However, most existing research efforts on recommendation were focused better serving overlooked the purpose item-providers. This paper is devoted to improve item exposure fairness for item-providers’ objective, keep accuracy not decreased or even improved users’ objective. We propose set stock volume constraints items, be specific, limit maximally allowable recommended times an proportional frequency its being interacted past, which validated achieve superior common recommenders thus mitigates Matthew Effect popularity. With pre-existing length our volumes a heuristic strategy based normalized scores Minimum Cost Maximum Flow (MCMF) model proposed solve optimal user-item matching problem, whose performances than that baseline algorithm regular context, line with state-of-the-art enhancement baseline. What’s more, MCMF parameter-free, while those counterpart algorithms have resort parameter traversal process their best performance.